A Power-Efficient Approach to Adaptive Polling ... - IEEE Xplore
484 IEEE COMMUNICATIONS LETTERS, VOL. 11, NO. 6, JUNE 2007 2) The poll is not received at station k, k does not respond to the AP and the AP proceeds to poll the next station at time t + t POLL +4∗ t PROP DELAY + t BUFF DATA + t DATA +t ACK . In this case the AP increases the choice probability of station v with the highest current reward estimate d v and then lowers the value of the reward estimate d k of station k. In this case again b(j)=1. From the above discussion, it is obvious that the learning algorithm takes into account both the bursty nature of the traffic and the bursty appearance of errors over the wireless medium. The CP R−P algorithm is employed after each poll and is described in Figure 1. N is the number of mobile stations, L is the learning speed parameter, 0
NICOPOLITIDIS et al.: A POWER-EFFICIENT APPROACH TO ADAPTIVE POLLING PROTOCOLS FOR WIRELESS LANS 485 Delay (slots) 14 12 10 8 6 4 2 LPOAP RAP GRAP GRAPO Network N2 0 0 0.1 0.2 0.3 0.4 0.5 0.6 Throughput (packets/slot) Fig. 3. The delay versus throughput characteristics of LPOAP, RAP, GRAP and GRAPO when applied to network N2. Average power consumption (Watts) 2 1.8 1.6 1.4 1.2 1 0.8 0.6 Network N1 LPOAP−no power saving LPOAP−low power mode Average power consumption (Watts) 2 1.8 1.6 1.4 1.2 1 0.8 0.6 Network N2 LPOAP−no power saving LPOAP−low power mode such cases the increased number of collisions a) reduces the throughput of RAP at high loads and b) at the same time the average delay is also increased a lot . These two remarks explain the curve of RAP at high loads in Figure 2. Finally, under heavy bursty traffic conditions (Figure 3) the number of active stations per polling cycle is significantly less than P RAP , resulting in increased overhead per DATA packet for RAP, GRAP and GRAPO when compared to LPOAP. 2) Increased power efficiency of LPOAP. When LPOAP utilizes the BUFF DATA control packet to inform mobile stations to go to the DOZE power state, its average power consumption is significantly reduced. This can be seen from Figure 4. These power savings increase for an increasing load as in such conditions, the learning mechanism almost always polls mobile stations that have a buffered packet, a fact that leads the other mobile stations to go to the DOZE state. Contrary, in low offered loads, polled stations are most of the time idle and thus the other mobile stations seldomly go to the DOZE state. Moreover, we see more power gains in N1 than N2 because in networks with more mobile stations (N1), each data packet transmission from a polled mobile station results to a transition to the DOZE state of a larger percentage of mobile stations. Finally, the low-power operation does not impact the performance of LPOAP, as the results in Figures 2, 3 were obtained for the low-power mode of LPOAP and are identical to those we obtained for LEAP. This is a positive property as although it is indeed desirable it is not always possible for MAC protocols . 0.4 0.2 0 Fig. 4. 0.2 0.4 0.6 0.8 1 Offered Load (packets/slot) Average mobile power consumption for LPOAP. 0.4 0.2 0 0.2 0.4 0.6 0.8 1 Offered Load (packets/slot) for a DATA packet transmission. LPOAP requires an overhead of three control packets per DATA packet (POLL, BUFF DATA, ACK). RAP, GRAP and GRAPO can transmit at most five DATA packets per with an overhead of sixteen control packets, for L RAP =1, (READY, CDMA transmission of random addresses which is equal to five times the duration of a control packet, five POLL packets, five ACK packets) yielding an overhead of 3.2 control packets per DATA packet. However, this scenario seldomly occurs due to the increasing number of collisions in RAP when the number of active stations per polling cycle approaches the number of random addresses, P RAP . This happens in high loads and in IV. CONCLUSION This paper proposed the LPOAP MAC protocol for infrastructure WLANs. It operates efficiently under burtsy traffic and can reduce mobile power consumption without performance penalties. REFERENCES  P. Nicopolitidis, G. I. Papadimitriou, and A. S. Pomportsis, “Learningautomata-based polling protocols for wireless LANs,” IEEE Trans. Commun., vol. 51, no. 3, pp. 453-463, March 2003.  M. A. L. Thathachar and P. S. Sastry, “Estimator algorithms for learning automata,” in Proc. Platinum Jubilee Conference on Systems and Signal Processing, Bangalore, India, Dec. 1986.  K. C.Chen, “Medium access control of wireless LANs for mobile computing,” IEEE Network, pp. 50-63, Sept./Oct. 1994.  M.-C. Li and K.-C. Chen, “GRAPO-optimized group randomly addressed polling for wireless data networks,” in Proc. 44th IEEE VTC, June 1994, pp. 1425-1429.  T. D. Lagkas, G. I. Papadimitriou, P. Nicopolitidis, and A. S. Pomportsis, “A new approach to the design of MAC protocols for wireless LANs: combining QoS guarantee with power saving,” IEEE Commun. Lett., vol. 10, no. 7, pp. 537-539, July 2006.  Y.-K. Sun and K.-C. Chen, “Energy-efficient multiple access protocol design,” IEEE Commun. Lett., vol. 2, no. 8, pp.334-336, Dec. 1998.